Handle NoType in TypeComparer.disjointnessBoundary

[wjoel]

Closes #20265

Note that the example in that issue already does not compile on 3.5.0 (and 3.4.3), failing with a reasonable error message. However, opting in to ignoring the issue with -source:3.3 does cause a crash.

Compiling project (Scala 3.5.0, JVM (17))
[error] ./Main.scala:8:27
[error] The match type contains an illegal case:
[error]     case k1 + k2 --> v => Hinze[k1 --> v] * Hinze[k2 --> v]
[error] (this error can be ignored for now with `-source:3.3`)
[error] ./Main.scala:22:96
[error] Cannot prove that (String --> Unit) * (String --> Unit) =:= Hinze[String + String --> Unit].
[error]   println(summon[((String --> Unit) * (String --> Unit)) =:= Hinze[(String + String) --> Unit]])
[error]                                                                                                ^
[error] ./Main.scala:23:67
[error] Cannot prove that String =:= Hinze[Fix[[X] =>> String + String * X + X * X] --> Unit].
[error]   println(summon[String =:= Hinze[Fix[Lambda[String]#L] --> Unit]])
[error] 

[wjoel]

defn.NothingType is another alternative, but using AnyType gives error messages similar to what we get from Scala 3.3.

[sjrd]

defn.AnyType is definitely better. NoType shouldn't be here. At best we have to treat it like ErrorType just above.

Perhaps add an assertion that we only get here under -source 3.3? Like

assert(!sourceVersion.isAtLeast(SourceVersion.`3.4`),
    "Unexpected NoType found in disjointnessBoundary")

[eejbyfeldt]

I believe the example I added to #20265

trait Expr[T]

trait Ops[T](using m: scala.deriving.Mirror.ProductOf[T]) {
  def apply(args: Tuple.Map[m.MirroredElemTypes, Expr]): Expr[T] = ???
}

case class P(a: Int)
object P extends Ops[P]

hits this crash without specifying -source 3.3.

[sjrd]

Then it should also be added as a test.

[wjoel]

@sjrd I've added it as tests/neg/i20265-1.scala

[eejbyfeldt]

I found an added another piece of code hitting this crash in the issue #20265 Maybe it worth revisiting this fix? It also seems to fix my issue.

[wjoel]

I see @sjrd is the assignee, maybe he has some thoughts on this? Not sure how to bring this further.